The Improvement Of Neuro-Expert Method For Anomaly Detection In Nuclear Reactor

THE IMPROVEMENT OF NEURO-EXPERT METHOD FOR ANOMALY DETECTION IN NUCLEAR REACTOR

Muhammad Subekti
PTRKN - BATAN, Puspiptek, Serpong Tangerang, Indonesia
ABSTRACT
The improvement of Neuro-Expert Method have been done for anomaly detection in nuclear reactor by utilization of Recurrent Neural Network (RNN). The development of Multilayer Perceptron (MLP) for similar objective was done in the previous research. Due to the monitoring system needs of redundancy to assure the reactor safety, the Neuro-Expert have been improved in this research by utilization of RNN as added method. Furthermore, the expert system is coupled to MLP and RNN by using specific parameters which depend on the training characteristic. The offline demonstation of MLP and RNN were carry out for PWR Borselle, simulator of PWR Surry-1, RSG-GAS reactor, and High Temperature Engineering Tested Reactor (HTTR). The learning results showed unsignificant different of maximum error of 0.0061 for MLP and 0.0049 for Jordan typed RNN. In the contrary, Elman typed RNN gave unreliable maximum error of 0.0130.

Keywords: improvement, Neuro-Expert, multilayer perceptron, Recurrent Neural Network, anomaly detection.
Proceeding Seminar Nasional Ke-15 "TEKNOLOGI DAN KESELAMATAN PLTN SERTA FASILITAS NUKLIR", Surakarta, 17 Oktober 2009